Former Fed Chair Bernanke Appointed to Anthropic's Long-Term Benefit Trust

On July 9, Anthropic announced that Dr. Ben Bernanke, former Chair of the Federal Reserve Board (FRB), will join the company's Long-Term Benefit Trust (LTBT) as a new member.

The LTBT is an independent body that oversees Anthropic's adherence to its mission of "responsible development of advanced AI for the long-term benefit of humanity." Bernanke served as FRB Chair from 2006 to 2014, steering the institution through the 2008 global financial crisis and leading the subsequent recovery. After his economics research career at Princeton, he was awarded the Nobel Memorial Prize in Economic Sciences in 2022.

"AI has enormous potential, but the range of outcomes is just as wide," Bernanke said. "How that potential is realized depends in part on the institutions we build around it." Having someone with experience managing an economic crisis join AI governance adds new weight to the governance structures of AI companies.

Meta's "Muse Spark 1.1" Enters the Price War — $4.25 per Million Output Tokens

Meta is making a full-scale entry into the AI API market. Its new model, Muse Spark 1.1, is priced at $4.25 per million output tokens — undercutting Grok 4.5, which was released the day before.

Compared with API pricing from OpenAI and Anthropic, this is exceptionally cheap. As the AI price war intensifies, Meta appears to be aiming to win API market share with low pricing as its weapon. That said, a low price does not necessarily mean low quality — this kind of cost efficiency has the effect of pulling down price levels across the entire industry.

While welcome news for API users, the move puts revenue pressure on competitors. The race to cut AI inference costs looks set to accelerate even further.

poolside's "Laguna XS 2.1" — A 33B MoE Coding Model That Runs Locally

Startup poolside has released Laguna XS 2.1, a model built for coding agents. It adopts a Mixture-of-Experts (MoE) architecture with 33B total parameters, of which only 3B are active per token.

The model reportedly achieves a +5.4% improvement over the previous version on SWE-bench Multilingual, with performance gains on terminal-format tasks as well. It is designed to run on local machines and is also available via OpenRouter.

This approach — achieving strong coding performance with a small parameter count — looks promising for enterprise adoption from privacy and latency perspectives.

Dartmouth Study: AI Errors Can Actually Slow Physicians Down in Message Writing

A Dartmouth study suggests that when AI is used to write patient-facing messages, AI errors may end up consuming more of physicians' time.

In some cases, the process of reviewing and correcting inaccuracies and awkward phrasing in AI-generated text takes longer than writing from scratch. This is an important finding, showing that AI adoption does not always translate into efficiency gains.

AI deployment in healthcare requires careful evaluation that goes beyond simple "time savings," and the study underscores that the reliability of AI output directly affects workflow efficiency.

AI-Generated Content Floods Social Media — LinkedIn Hit Hardest

A Pangram investigation reports that AI-generated content on social media is surging, with the trend especially pronounced on LinkedIn. On Hacker News, the topic drew 79 points and 56 comments, sparking active community discussion.

While the spread of AI tools lowers the barrier to content creation, concerns are growing about declining information quality on these platforms. The difficulty of telling what is human-made and what is AI-generated presents a new challenge: content trust.

Takeaways

Today's news paints a picture of AI governance becoming institutionalized even as price wars and model downsizing unfold simultaneously in the market. Bernanke joining Anthropic is a signal of growing external weight on AI company governance. Meta's aggressive pricing is a boon for users, but it raises questions about the industry's revenue structure as a whole. And the realities on the ground in healthcare and social media reflect a world where practical AI adoption cannot be judged on speed and price alone.

---